Visitors to the fifth floor of Anthropic's San Francisco headquarters are greeted by warm wood, soft light, a portrait of Alan Turing, and a receptionist who hands them a small book, the size of the pocket Bibles proselytizers distribute on street corners. It's a copy of Machines of Loving Grace, a 14,000-word essay CEO Dario Amodei wrote in 2024, laying out a utopian vision for how AI could transform the world by accelerating scientific discovery. In January he published a second novella-length essay, The Adolescence of Technology, detailing the attendant dangers: mass surveillance, widespread job losses, even permanent loss of human control.
That's how Anthropic introduces itself to strangers. Not with a brochure or a pitch deck, but with a CEO's intellectual reckoning with the thing his company is building.
Most coverage of Anthropic right now is consumed with the Pentagon confrontation: the supply-chain-risk designation, the lawsuits, Trump calling them "radical left nut jobs." It's genuinely dramatic, and I'll get to it. But the confrontation is, in a sense, a symptom. What produced it is the more interesting story, and the one with lessons that extend well beyond AI.
What the Culture Actually Looks Like
Anthropic was founded in 2021 by Dario and Daniela Amodei and five co-founders, all of whom had left OpenAI out of concern that the organization was moving too fast. Before they had a product, they built a "societal impacts" team. They hired a full-time in-house philosopher, Amanda Askell, whose job is to shape Claude's moral sensibilities and prepare it for a future where it may be more intelligent than its creators. "It does sometimes feel a little bit like you have a 6-year-old, and you're teaching the 6-year-old what goodness is," Askell told Time. "By the time they're 15, they're going to be smarter than you at everything."
Employees call themselves "ants." Many maintain a digital notebook, a Slack channel where they post stream-of-consciousness thoughts, hopes, fears, and observations. Dario writes his own lengthy entries. He also gives biweekly company-wide lectures called "Dario Vision Quests" (a name he tried to resist, for what should be obvious reasons). The format is deliberately unfiltered: where's the company, what does he believe, what worries him. No corporate-speak, no defensive positioning. The goal, he has said publicly, is simply "to get a reputation for telling the company the truth about what's happening, to call things what they are, to acknowledge problems." That sounds basic. Most organizations at scale never actually do it.
He reportedly spends roughly 40% of his time on culture, not product or technology. For a company now valued at $380 billion with 2,500 employees, that's a striking allocation. Most executives I've known would answer "culture" if asked what matters most, then spend 40% of their time in budget reviews.
The cultural filtering starts before anyone joins. Every candidate, regardless of role, goes through a universal culture interview designed to assess genuine values alignment, not just competence. Only employees with at least 30 days of tenure, who've completed multi-stage training on Anthropic's culture, are allowed to conduct those interviews. The logic is that cultural transmission is too important to delegate to someone still figuring out what the culture is. A sample interview question, cited in the Time profile: Would you be willing to lose the value of your stock if Anthropic decides not to release models because it can't guarantee they're safe? That's not a hypothetical. It's a filter. They will walk away from technically brilliant candidates who can't answer it honestly.
Every technical employee, from new hire to founding executive, carries the same title: Member of Technical Staff. No senior vice presidents, no distinguished fellows, no hierarchy signaling. Authority comes from the quality of your argument, not your position on an org chart. Former Google employee Daniel Freeman, now on Anthropic's red team, put it plainly: Anthropic's competitors contain fiefs "that all care about different things and are low-key at war with each other. I've absolutely never felt that at Anthropic."
The result is an 80% retention rate and offer acceptance rates approaching 95% for some roles, in a market where competing offers arrive constantly. Work-life balance scores lowest in reviews, and the intensity is real. One employee review on Blind captured the internal atmosphere well: "If you work at Anthropic you may become convinced that you are working at the most important thing, at the most important time, at the most important company, perhaps in all of history. This may or may not be true, at the very least, take it with a grain of salt." The self-awareness embedded in that caveat is unusual. Most mission-driven organizations don't produce it.
The Test
All seven co-founders have pledged to give away 80% of their wealth. Anthropic is incorporated as a public benefit corporation, legally bound to balance private and public interests. The EA (effective altruism) movement, which emphasizes using reason to do the most good and averting catastrophic risk, runs through the company's DNA, though the Amodeis are careful about the label given its associations with Sam Bankman-Fried.
These aren't incidental details. They explain why, when the Pentagon demanded Anthropic grant unrestricted access to Claude for "all lawful purposes," including the mass surveillance of American citizens and fully autonomous weapons systems, Amodei said no, and held that position when it cost him a $200 million contract and triggered a designation as a national security supply-chain risk. There's a version of that negotiation in which a company with different cultural foundations folds. Anthropic didn't.
The reverberations were significant. Claude shot to number one on the App Store the day after the blacklisting, displacing ChatGPT. More than 30 employees from OpenAI and Google DeepMind, including Google Chief Scientist Jeff Dean, filed an amicus brief supporting Anthropic in court, signing in their personal capacities. That has apparently never happened between competing AI companies. OpenAI's robotics team lead resigned over her company's rapid pivot to a Pentagon deal hours after Anthropic was blacklisted. A senior OpenAI researcher announced he was moving to Anthropic. Hundreds of employees from both companies signed an open letter urging their executives to stand with Anthropic's position.
What propagated wasn't just news coverage. It was the demonstration that a culture genuinely built around a principle can function as a recruiting and retention advantage in ways that compensation packages cannot replicate. People who care about these questions now know exactly where to apply.
The Real Risks
A genuinely honest assessment includes the caveats, and Anthropic is not short of them.
The first risk is insularity. Deep, shared belief systems produce extraordinary commitment and equally extraordinary blind spots. The head of Anthropic's Safeguards Research Team recently resigned, writing that "throughout my time here, I've repeatedly seen how hard it is to truly let our values govern our actions," and that employees "constantly face pressures to set aside what matters most." The internal tension that produces is worth taking seriously, especially in a company whose societal impacts team leader told Time, regarding AI-driven job displacement: "It feels like we might be speaking out of both sides of our mouths."
The second risk is scalability. Maintaining cultural coherence at 2,500 people requires the CEO to spend 40% of his time on it. That math doesn't hold at 25,000. Most organizations that tried to scale a strong founding culture have experienced steep, fast decay.
The third risk Anthropic has already started to confront: mission drift under competitive pressure. The same week Anthropic stared down the Pentagon, it quietly rewrote its Responsible Scaling Policy, dropping a binding commitment to pause development if it couldn't guarantee safety adequacy. Chief science officer Jared Kaplan told Time it was "naive" to think they could identify bright lines between danger and safety, and that it made no sense to make unilateral commitments "if competitors are blazing ahead." Critics noted the timing. The pattern where foundational commitments get reinterpreted just enough to accommodate what you were going to do anyway is the quiet death of most cultures that don't survive their own success. Dave Orr, Anthropic's head of safeguards, may have said it best: "We're driving down a cliff road. A mistake will kill you. Now we're driving at 75 instead of 25."
What Traditional Companies Can Actually Use
So what transfers to organizations that aren't building AI, aren't worth $380 billion, and aren't fighting Pete Hegseth?
Culture functions as a management system, not a motivational poster. If the CEO isn't investing real, consistent time in it, it doesn't operate as a functional mechanism. It exists as a document on an intranet that nobody reads after orientation. The DVQ concept, a regular unfiltered communication from leadership about what's actually true in the organization, is something any company can implement and almost none do well at scale. Publishing your thinking publicly, as leaders, creates accountability that press releases and earnings calls don't.
The hiring filter matters more than the mission statement. Selective hiring that actually turns away excellent candidates on cultural grounds is universally endorsed in principle and universally abandoned in practice when there's a role to fill and a manager under pressure. Building a process designed to survive that pressure, with trained interviewers and a consistent culture screen for every role, is an engineering problem as much as a values problem. And asking candidates whether they'd sacrifice something real for the mission, not hypothetically but as an actual question with actual weight, tells you things behavioral interviews don't.
Values that have never been tested haven't been proven. The real signal of a healthy culture isn't the annual engagement survey; it's what decisions get made when holding the line costs you something meaningful. Most organizations will never face a moment as stark as Anthropic's. But the question is worth asking before circumstances force it: if your company's stated values were genuinely put to the test, what would the decision look like?
A senior Pentagon official reportedly said that disentangling from Claude would be "an enormous pain in the ass." Anthropic said fine. That kind of clarity doesn't appear in a company overnight. It's built through how you hire, what your leaders spend their time on, and whether you've ever actually decided what you won't do.